Evolution in Swarm Intelligence: An Evolutionary Ant-Based Optimization Algorithm

Swarm Intelligent (SI) algorithms draw their inspiration from the interaction of individuals of social organisms. One such algorithm, Ant Colony Optimization (ACO) [1], utilizes the foraging behavior of ants to solve combinatorial optimization problems. Although ACO performs well in a static environment, it has been pointed out that ACO does not perform as well as other heuristics in dynamic situations such as routing. This paper proposes a new algorithm, entitled Evolutionary Ant Colony Optimization (EACO), that combines ACO with elements of Genetic Algorithms (GA). By adding evolution, the EACO algorithm allows the individual ants to develop their own characteristics, thereby removing the homogeneity inherent within ACO. Our results demonstrate the potential of this approach in a dynamic environment. There have been other attempts in the past at combining SI and Evolution, however, most have used the two as discrete events. White et al.[2] came the closest to mimicking nature with their approach by moving two of the global ACO parameters to the individual ants and evolving the ants themselves. However, even in White’s experiments, the two algorithms (GA and ACO) are run as discrete steps where the GA runs only after all ants have found a solution. EACO attempts to approximate nature by directly integrating the bio-operators of GA into ACO and evolving the ants in the system on a continuous basis. In EACO, an initial population of ants is created, and as ants die and are removed from the system, new ants are created in their place. This approach allows the algorithm to run continuously as a single system rather than as two discrete events where entire populations are created and destroyed between each iteration of the GA. The ants in the ACO algorithm proposed by White et al.are characterized by two main characteristics: sensitivity to pheremone and sensitivity to link cost. On the other hand, the genotype in EACO contains five attributes: lifespan, pheremone quality, pheremone sensitivity, speed, and reproduction rate. Our current implementation uses only lifespan and speed. Two different sets of tests were used to test the efficacy of the EACO algorithm versus traditional ACO. In both tests, a simple map was used to test the ability of each algorithm to find the optimal path between two cities. In the first test, a search for the shortest path was done by both algorithms in a purely